2 ms·
Well, of course. Large language models reflect the average of the input data.
by Animats 3y ago
Well, of course. Large language models reflect the average of the input data.
- astrange 3y agoAn average of the input data would be a scalar value, not a few GB model. Don't go around making claims about how a instruct-tuned text model works if you've never tried to operate a foundation model. It's very obvious those two aren't doing the same kind of thing; they can't both be "average text". And of course don't confuse the sampler for the model; you can change those out.